Scholarlink

<p>Remote sensing image analysis plays a vital role in geographic information science, yet faces significant challenges due to massive data volumes, high dimensionality, heterogeneity, and noise interference. Traditional manual or semi-automated methods often struggle with inefficiency and limited accuracy, while existing deep learning-based approaches are hindered by subjective labeling, multi-source inconsistencies, and constrained generalization capabilities. This paper propose GrsPT, a novel adversarial learning framework for remote sensing image understanding that integrates two large language model (LLM)-powered agents: GenPT for image analysis and generation, and CheckPT for content quality evaluation. GrsPT introduces three evaluation metrics: geometric accuracy, spectral consistency, and temporal dynamics, to systematically assess and refine generated outputs. By employing an iterative prompt optimization mechanism guided by the discriminator, the framework progressively enhances output quality against predefined standards. Experimental results demonstrate that GrsPT achieves superior accuracy, consistency, and authenticity compared to existing methods, offering scalable generalization and establishing a valuable benchmark for future research in remote sensing image analysis.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

GrsPT: an adversarial LLM-driven framework for enhanced remote sensing image analysis with multi-dimensional evaluation

  • 期刊论文

  • 发表日期：

    2026年5月6日

查看全文
  • Bing Li,
  • Xiaoyan Yang,
  • Tao Shi,
  • Lingtong Meng,
  • Chenhao Zhao

摘要

Remote sensing image analysis plays a vital role in geographic information science, yet faces significant challenges due to massive data volumes, high dimensionality, heterogeneity, and noise interference. Traditional manual or semi-automated methods often struggle with inefficiency and limited accuracy, while existing deep learning-based approaches are hindered by subjective labeling, multi-source inconsistencies, and constrained generalization capabilities. This paper propose GrsPT, a novel adversarial learning framework for remote sensing image understanding that integrates two large language model (LLM)-powered agents: GenPT for image analysis and generation, and CheckPT for content quality evaluation. GrsPT introduces three evaluation metrics: geometric accuracy, spectral consistency, and temporal dynamics, to systematically assess and refine generated outputs. By employing an iterative prompt optimization mechanism guided by the discriminator, the framework progressively enhances output quality against predefined standards. Experimental results demonstrate that GrsPT achieves superior accuracy, consistency, and authenticity compared to existing methods, offering scalable generalization and establishing a valuable benchmark for future research in remote sensing image analysis.